BigQuery 中的 ADK 智能体分析

---
name: adk-agent-analytics-in-bigquery
description: "Inspect an ADK analytics table and query bounded production-agent behavior."
---

# ADK Agent Analytics in BigQuery

Inspect an ADK analytics table and query bounded production-agent behavior.

## Required Aident Actions

- <action-tag>composio:googlebigquery_tools:googlebigquery_get_table_schema</action-tag> (required inputs: project_id, dataset_id, table_id): Tool to fetch a BigQuery table's schema and metadata without querying row data. Use before generating SQL queries to avoid column name typos and confirm field types and nullable modes. This is especially useful when...
- <action-tag>composio:googlebigquery_tools:googlebigquery_query</action-tag> (required inputs: project_id, query): Query Tool runs a SQL query in BigQuery using the REST API. Use proper BigQuery SQL syntax, e.g., SELECT * FROM `project.dataset.table` WHERE column_name = 'value'. Results are returned under data.rows; an empty rows...

Use Aident Loadout to read the current Action schema before constructing inputs. Check the required integration connection in Aident Vault. If an Action is billable or mutating, run Aident preflight, show the affected target and quoted cost or risk, and wait for explicit user confirmation before execution.

## Workflow

1. Confirm the requested outcome, source material, destination, audience, constraints, and acceptance criteria.
2. Apply the source-derived guidance below to create a concrete plan. Resolve ambiguity before invoking an Action.
3. Select only the Aident Actions whose documented effect directly advances the requested outcome. Do not invoke every listed Action by default.
4. Inspect the current schema and prepare the minimum valid input for each selected Action.
5. Preflight each selected Action. Execute it only after any required confirmation, in dependency order, and carry returned IDs or asset URLs into later steps.
6. Verify the returned IDs, URLs, statuses, or artifacts against the acceptance criteria. Report partial completion precisely and do not repeat paid or mutating calls blindly.

## Source-Derived Guidance

- Verify the exported table schema and identify privacy-sensitive prompt or response fields.
- Use aggregate queries where possible and avoid returning raw sensitive payloads.

## Execution Boundaries

- Treat upstream provider-specific commands as background knowledge only. Execute the workflow through the exact Aident Actions above.
- Never request raw credentials in chat. Use Aident Vault connection flows for required accounts.
- Preserve user-provided wording, brand constraints, rights restrictions, and target identifiers. Do not invent authorization.
- Read current state before a mutation, state the exact target, and include a rollback or recovery step in the result.
- Stop when a required integration is disconnected, preflight rejects the input, the target is ambiguous, or the user declines a required confirmation.

## Output

Return the execution plan, selected Action names, confirmed targets, Aident result identifiers or asset URLs, verification evidence, and any remaining blocked step.

## Attribution

This Skill adapts the reviewed upstream workflow. See [UPSTREAM.md](UPSTREAM.md) for the pinned source and [LICENSE.txt](LICENSE.txt) for the preserved license.